Re: [PATCH 2/2] perf unwind: fix libunwind when tid != pid

From: John Keeping
Date: Sun Aug 04 2019 - 07:45:20 EST


On Fri, 2 Aug 2019 15:30:39 +0200
Jiri Olsa <jolsa@xxxxxxxxxx> wrote:

> On Mon, Jul 29, 2019 at 06:24:30PM +0100, John Keeping wrote:
> > Commit e5adfc3e7e77 ("perf map: Synthesize maps only for thread group
> > leader") changed the recording side so that we no longer get mmap events
> > for threads other than the thread group leader.
> >
> > When a file recorded after this change is loaded, the lack of mmap
> > records mean that unwinding is not set up for any other threads.
>
> sry I dont' follow what's the problem here, could you please
> describe the scenrio where the current code is failing in
> more details

With perf compiled to use libunwind, run:

perf record --call-graph=dwarf -t $TID -- sleep 5
perf report

If $TID is a process, then the output has the call graph, but if it's a
secondary thread then it is as if --no-call-graph was specified.

> >
> > Following the rationale in that commit, move the libunwind fields into
> > struct map_groups and update the libunwind functions to take this
> > instead of the struct thread. This is only required for
> > unwind__finish_access which must now be called from map_groups__delete
> > and the others are changed for symmetry.
> >
> > Note that unwind__get_entries keeps the thread argument since it is
> > required for symbol lookup and the libdw unwind provider uses the thread
> > ID.
>
> SNIP
>
> > @@ -59,37 +59,31 @@ int unwind__prepare_access(struct thread *thread, struct map *map,
> > return 0;
> > }
> > out_register:
> > - unwind__register_ops(thread, ops);
> > + unwind__register_ops(mg, ops);
> >
> > - err = thread->unwind_libunwind_ops->prepare_access(thread);
> > + err = mg->unwind_libunwind_ops->prepare_access(mg);
> > if (initialized)
> > *initialized = err ? false : true;
> > return err;
> > }
> >
> > -void unwind__flush_access(struct thread *thread)
> > +void unwind__flush_access(struct map_groups *mg)
> > {
> > - if (!dwarf_callchain_users)
> > - return;
>
> why did you remove this check?

I don't think there is any way for unwind_libunwind_ops to be set if
!dwarf_callchain_users so this is redundant given the following
condition.

But this should probably be a separate patch.

> > -
> > - if (thread->unwind_libunwind_ops)
> > - thread->unwind_libunwind_ops->flush_access(thread);
> > + if (mg->unwind_libunwind_ops)
> > + mg->unwind_libunwind_ops->flush_access(mg);
> > }
> >
> > -void unwind__finish_access(struct thread *thread)
> > +void unwind__finish_access(struct map_groups *mg)
> > {
> > - if (!dwarf_callchain_users)
> > - return;
>
> why did you remove this check?

Likewise.


Regards,
John